Add Grayscale image preprocessor

[zaleslaw]

The Grayscale image preprocessor is missed in KotlinDL.
Convert image to grayscale.

The main question is: should we keep the ImageShape with 3 channels or with 1 channel? Or support the difference via parameter like in PyTorch

The desired PR addressing this issue should include:

• Implementation (you can take inspiration from the implementation of Cropping as reference)
• Documentation
• JUnit tests in dataset module
• An example with usage of this image preprocessor

P.S. If you want to take this ticket, please leave the comment below
P.P.S Read the Contributing Guidelines.

The reference implementation could be taken from torchvision.transforms

[zaleslaw]

@juliabeliaeva could you have a look at this ticket

[juliabeliaeva]

@zaleslaw yes, assign this issue to me, please.

[juliabeliaeva]

should we keep the ImageShape with 3 channels or with 1 channel?

Since currently BufferredImage is used for image preprocessing, in my PR I use BufferedImage.TYPE_BYTE_GRAY and return images with one channel. If we migrate to something else, we could think about supporting 3 channels for grayscale images.